As one of the leading property managers in Bali, Bukit Vista is constantly finding new innovations to inspire delight for our partners, guests, and people within Bukit Vista. One of the countless innovations we implemented is using AI for our reservation and hosting team for improved workflow. Rafie and Rifqu, one of our data scientist interns are able to implement this through OSAI (Operations Supervision Artificial Intelligence) Read more below on how they use AI to inspire delight for our guests and operations team at Bukit Vista.
Rafie is majoring in data science at Airlangga University while Rifqi is majoring in engineering at Universitas Diponegoro , they are both data scientist interns here at Bukit Vista. Their experience with data has brought them to learn more about machine learning. With experience in making models using Python, visualizing data with Data Studio, developing with TensorFlow, and working with databases through MariaDB. And are here to explain more about their project during at Bukit Vista.
A background of the operation work at Bukit Vista
Imagine you have a fast-paced company with a lot of people coming and going, and a lot of information that needs to be digested by new people. The leaders, HR, and seniors of your company would be tired of teaching new people over and over again about the essential things in your company, right? This has been one of the big problems Bukit Vista faced, especially in the Guest Squad/Hospitality, Air support, and Reservation (HAR) Team. Transferring knowledge has been a problem for the HAR team. Over time, the candidates change and the supervisors need to re-transfer their knowledge to the candidates. This repetitive and tiring process has caused the performance of the operations team to become somewhat seasonal. When new candidates come in, performance goes down, but over time, as the candidates learn and adapt, it goes back up. So, how can we keep performance stable while at the same time easing the supervisors’ job? The answer is a knowledge management system.
A knowledge management system is a tool or set of tools that organizations use to store, organize, and access information.The goal of a knowledge management system is to make it easier for people within an organization to find and use the information they need to do their jobs effectively. This can include things like documents, policies, procedures, and expertise that are relevant to the organization’s operations.
After a week of research to define the best method to solve this issue, we decided to use BERT, specifically Roberta-base-squad2, to build our knowledge management system. BERT (Bidirectional Encoder Representations from Transformers) is a powerful language processing model developed by Google. It is often used for tasks such as natural language processing and text classification. One popular application of BERT is in question answering, where the model is trained to answer questions based on a given input text.
Roberta-base-squad2 is a variation of the BERT model that is specifically designed for the question answering task. It is an improved version of BERT that was developed by researchers at the Allen Institute for Artificial Intelligence. It uses a technique called transfer learning, where the model is pre-trained on a large dataset and then fine-tuned for the specific task at hand. This allows the model to learn general language patterns and then quickly adapt to the question answering task, giving it a significant performance boost.
Cool, right? Enough with the technical stuff. In simple terms, BERT, specifically Roberta-base-squad2, is a chatbot that you can personalize using your own company data or information. We at Bukit Vista then used Roberta-base-squad2 using our own data, which consists of more than 150 listing descriptions, amenities, and detailed property data.
Here is the result of our newly-built tool, which we called OSAI (Operations Supervision Artificial Intelligence):
OSAI is able to cut the time our employees used to spend looking for information, which could range from three minutes to one hour (depending on whether that information was available on the listing or not), down to less than 30 seconds. Amazing, right? Another reason why we love OSAI is that it’s like a buddy that you can teach. So, when OSAI is not able to answer the questions that people bring up, it’s mainly because we don’t have that information stored in our database yet. We can then easily store the newly-found information on our website that everybody on BV can access, and voila! OSAI will be able to answer those kinds of questions the next time. So, how does OSAI work?
When the user inputs a question and the unit name, OSAI will search for specific unit data, including the listing description, amenities, and additional property information. This information will be used as the context for OSAI to generate answers to the user’s question. OSAI uses its transformer architecture to process language and understand the relationships between words in the question (semantic analysis). After processing the question, OSAI will search the fetched context for the best answer to the question.
What do they say about OSAI?
Overall rating : 8.17/10
The rating speaks for itself. Overall, the users were satisfied with the OSAI ability to answer the inquiry in seconds so they won’t need to spend more time looking for answers. However, one big concern still put in place; the ability of OSAI to give the relevant answer. This issue happens because there were no related data uploaded to the database making OSAI not able to give the relevant answer. No worries though, the developers have noticed this issue and they are on the way fixing it!
Here are some sneak peeks of the review from the users
“It still have some irrelevant answer with the questions that asked, still need improvement for that, but overall it really help”
“OSAI is really helpful for the daily operations of both the Reservation team and the Hosting team. The point that OSAI can answer any questions correctly when the data is available, gives BV employees a shortcut for an answer rather than searching anywhere including calling the property owner. OSAI also provides a solution for the recorded data that has been asked before. To give more data to the OSAI, is the main thing to improve OSAI.”
“I think it still needs a lot of improvement. My questions have not been answered correctly. We need to put a lot more data into this osai. But i believe overtime it will be fixed.”
Does that sound interesting? If so, come and join us here at Bukit Vista. We have a lot of cool things to try out here, or you can even build your own tools that inspire delight!